Maximum Usable Period of Ground Motion Records for Engineering Applications

Seismologists traditionally define the usable bandwidth of ground motion records based on signal-processing principles that ensure acceptable signal-to-noise ratios. Engineers rely on these records to assess the responses of structures whose fundamental periods, T, fall within the bandwidth. The upper bound of this interval, TUP, depends on the long-period cut-off of the filter applied to the accelerograms, TC. This study examines whether, and to what extent, ground motion filtering may lead to underestimating the response of systems with T  ≤ TUP, especially for structures experiencing period elongation in the post-elastic range. If a systematic bias exists, current TUP values may be overly permissive. To address this concern, we computed benchmark responses for single-degree-of-freedom (SDOF) systems with varying T to several high-quality records processed with very long cut-off periods. These ground motions were then incrementally filtered, and the SDOF systems’ responses reassessed. By systematically comparing the benchmark response estimates with those obtained from the degraded records, we identified an engineering-based upper bound period of the usable bandwidth of given ground motions, TUP-ENG. This threshold is such that the response bias of a structure with T  ≤ TUP-ENG remains below a predefined tolerance with certain statistical confidence, and varies based on whether the record is used for linear or nonlinear analyses of brittle or ductile structures. Our findings show that traditional TUP estimates are generally unconservative, i.e., TUP-ENG ≤ TUP, even for elastic analyses. This discrepancy becomes critical when predicting severe inelastic responses of systems with T near TUP, since the underestimation could be significant. To preserve structural response fidelity, we recommend adopting the application-specific engineering-based definition of the upper bound of the usable bandwidth of ground motions, which is more restrictive than the commonly accepted signal-processing-based single value.